[codex] document extract-knowhow redesign need

[ktwu01]

What changed

• document that extract-knowhow is still effectively Claude-first even though it is installed for both Claude Code and Codex CLI
• state that a broader redesign is likely needed if OpenScientist wants true multi-agent extraction and review
• outline the redesign direction: explicit peer review handoff, single-review contract, report-level review metadata, and a clear distinction between draft and reviewed skills

Why

The current package can install into multiple environments, but the workflow still behaves like a single-agent generation flow. That makes it too easy to treat extracted skills as reusable artifacts before defining how cross-agent review should work.

Impact

This PR does not change behavior. It makes the design gap explicit in the docs so future feature work has a clear target.